#!/bin/sh
# 部署或更新容器服务（Backend、Agent、Probe）。可以重复运行：每次都会拉取镜像、把配置迁移到当前版本需要的样子、
# 重新安装单元和更新器，然后重启节点；已有的身份和配置不会被覆盖。
#
#   首次部署 Agent：sudo sh deploy-node.sh agent --image-repo REPO --server https://panel --join jt_xxx --name hk-1
#   首次部署 Probe：sudo sh deploy-node.sh probe --image-repo REPO --server https://panel --join jt_xxx --name home-1
#   首次部署 Backend：sudo sh deploy-node.sh backend --channel dev [--public-url https://panel] [--port 29181] [--env KEY=VALUE]...
#     没有默认管理员：部署完打开面板，按页面引导自己创建管理员、注册 passkey、建立信任根；页面要的认领码在后端日志里（journalctl -u moshiro-backend）。
#   更新（配置迁移、单元和更新器、镜像）：sudo sh deploy-node.sh agent   （其余参数沿用上次的更新器配置）
#   旧版 Probe（probe.json 里没有 fleet 段）：加上 --server --join --name 重新运行即可，旧文件备份为 probe.json.old
#
# Backend 只支持 Podman，更新策略固定为 manual（管理端确认），监听宿主网络的 --port。
# 需要 root、systemd，以及 Podman 或 Docker；脚本要和 packaging/ 里的 systemd/、update/ 放在一起。
# rootless Podman 见 docs/deployment.md。
set -eu

usage() {
  echo "usage: deploy-node.sh <agent|probe|backend> [--engine podman|docker] [--image-repo R] [--channel dev|release] [--tag T] [--image REF]" >&2
  echo "       [--server URL --join TOKEN --name NAME] [--policy auto|manual] [--authfile F]" >&2
  echo "       backend: [--channel dev|release] [--port N] [--admin-user U --admin-password P：只给自动化用，正常不要] [--public-url URL] [--env KEY=VALUE]..." >&2
  exit 2
}

ROLE="${1:-}"; shift || true
case "$ROLE" in agent|probe|backend) ;; *) usage ;; esac
ENGINE=""; REPO=""; TAG=""; IMAGE=""; SERVER=""; JOIN=""; NAME=""; POLICY=""; AUTHFILE=""; CHANNEL=""; PORT=""; ADMIN_USER=""; ADMIN_PASSWORD=""; PUBLIC_URL=""; EXTRA_ENV=""
while [ $# -gt 0 ]; do
  case "$1" in
    --engine) ENGINE="$2"; shift 2 ;;
    --image-repo) REPO="$2"; shift 2 ;;
    --tag) TAG="$2"; shift 2 ;;
    --image) IMAGE="$2"; shift 2 ;;
    --server) SERVER="$2"; shift 2 ;;
    --join) JOIN="$2"; shift 2 ;;
    --name) NAME="$2"; shift 2 ;;
    --policy) POLICY="$2"; shift 2 ;;
    --authfile) AUTHFILE="$2"; shift 2 ;;
    --channel) CHANNEL="$2"; shift 2 ;;
    --port) PORT="$2"; shift 2 ;;
    --admin-user) ADMIN_USER="$2"; shift 2 ;;
    --admin-password) ADMIN_PASSWORD="$2"; shift 2 ;;
    --public-url) PUBLIC_URL="$2"; shift 2 ;;
    --env) EXTRA_ENV="$EXTRA_ENV$2
"; shift 2 ;;
    *) usage ;;
  esac
done
[ "$(id -u)" -eq 0 ] || { echo "run as root (rootless Podman: see docs/deployment.md)" >&2; exit 1; }

HERE="$(cd "$(dirname "$0")" && pwd)"
PACKAGING="$(cd "$HERE/.." && pwd)"
CONF_DIR=/etc/moshiro-update
CONF="$CONF_DIR/$ROLE.conf"

# 重复运行时沿用上次的更新器配置（仓库、策略、认证文件、引擎）
conf_get() { [ -r "$CONF" ] && sed -n "s/^$1='\\(.*\\)'\$/\\1/p" "$CONF" | head -n 1 || true; }
[ -n "$REPO" ] || REPO="$(conf_get IMAGE_REPO)"
[ -n "$POLICY" ] || POLICY="$(conf_get POLICY)"
[ -n "$AUTHFILE" ] || AUTHFILE="$(conf_get AUTHFILE)"
[ -n "$ENGINE" ] || ENGINE="$(conf_get ENGINE)"
[ -n "$CHANNEL" ] || CHANNEL="$(conf_get CHANNEL)"
if [ "$ROLE" = backend ]; then
  [ -n "$CHANNEL" ] || { echo "backend needs --channel dev or release" >&2; exit 2; }
  case "$CHANNEL" in dev) NAMESPACE=moshiro-dev ;; release) NAMESPACE=moshiro ;; *) echo "--channel must be dev or release" >&2; exit 2 ;; esac
  [ -n "$REPO" ] || REPO="registry.oeo.one/$NAMESPACE/moshiro-backend"
  [ -n "$PORT" ] || PORT=29181
  ENGINE=podman; POLICY=manual
fi
[ -n "$REPO" ] || REPO="registry.oeo.one/moshiro/moshiro-$ROLE"
[ -n "$POLICY" ] || POLICY=manual
if [ -z "$ENGINE" ]; then
  if command -v podman >/dev/null 2>&1; then ENGINE=podman; elif command -v docker >/dev/null 2>&1; then ENGINE=docker; else echo "podman or docker is required" >&2; exit 1; fi
fi
case "$ENGINE" in
  podman) UNIT_SRC="moshiro-$ROLE-container.service"; UNIT="$UNIT_SRC" ;;
  docker) UNIT_SRC="moshiro-$ROLE-docker.service"; UNIT="$UNIT_SRC" ;;
  *) echo "--engine must be podman or docker" >&2; exit 2 ;;
esac
V_CFG="moshiro-$ROLE-config"
V_STATUS="moshiro-$ROLE-status"
if [ "$ROLE" = backend ]; then UNIT=moshiro-backend.service; V_CFG=moshiro-backend-data; fi
for f in "$PACKAGING/systemd/$UNIT_SRC" "$PACKAGING/update/install-updater.sh" "$PACKAGING/update/moshiro-update.sh" "$PACKAGING/update/release-key.pub.pem"; do
  [ -r "$f" ] || { echo "missing $f (keep deploy-node.sh next to the packaging/ directory)" >&2; exit 1; }
done

# 1. 镜像：固定到 digest，后面的单元、辅助容器都用同一个引用。
#    默认从仓库里的签名清单取（<仓库>-manifest:<通道>，CI 发布的），用随脚本带的发布公钥验签，按清单里的 digest 固定——
#    仓库和网络都不需要可信，也不需要面板在线（Backend 第一次部署时还没有面板）。
#    --tag 指定标签就不验签（拉那个标签再固定 digest）；--image 直接用你给的引用。
if [ "$ENGINE" = docker ] && [ -n "$AUTHFILE" ]; then export DOCKER_CONFIG="$(dirname "$AUTHFILE")"; fi
pull_quiet() {
  if [ "$ENGINE" = podman ] && [ -n "$AUTHFILE" ]; then "$ENGINE" pull -q --authfile "$AUTHFILE" "$1" >/dev/null
  else "$ENGINE" pull -q "$1" >/dev/null; fi
}
resolve_verified() {
  channel="$CHANNEL"
  if [ -z "$channel" ]; then
    case "$REPO" in */moshiro-dev/*) channel=dev ;; */moshiro/*) channel=release ;; esac
  fi
  [ -n "$channel" ] || { echo "cannot tell the release channel from $REPO: pass --channel dev|release, or --tag to skip verification" >&2; return 1; }
  command -v openssl >/dev/null 2>&1 || { echo "openssl is required to verify the release signature (or pass --tag to skip verification)" >&2; return 1; }
  ref="$REPO-manifest:$channel"
  pull_quiet "$ref" || { echo "cannot pull the signed manifest $ref: is the repository readable (--authfile)? Or pass --tag to skip verification" >&2; return 1; }
  labels="$("$ENGINE" image inspect --format '{{index .Config.Labels "org.moshiro.manifest"}}|{{index .Config.Labels "org.moshiro.signature"}}' "$ref")" || return 1
  vtmp="$(mktemp -d)"
  printf '%s' "${labels%%|*}" | base64 -d > "$vtmp/manifest" 2>/dev/null && printf '%s' "${labels#*|}" | base64 -d > "$vtmp/sig" 2>/dev/null \
    || { rm -rf "$vtmp"; echo "the manifest in $ref is not readable" >&2; return 1; }
  if ! openssl dgst -sha256 -verify "$PACKAGING/update/release-key.pub.pem" -signature "$vtmp/sig" "$vtmp/manifest" >/dev/null 2>&1; then
    rm -rf "$vtmp"; echo "the signature of the manifest in $ref does not match the bundled release key; refusing to install" >&2; return 1
  fi
  mfield() { sed -n 's/.*"'"$1"'":"\([^"]*\)".*/\1/p' "$vtmp/manifest" | head -n 1; }
  vdigest="$(mfield digest)"; vcomponent="$(mfield component)"; vimage="$(mfield image)"; vtag="$(mfield tag)"
  rm -rf "$vtmp"
  [ "$vcomponent" = "$ROLE" ] && [ "$vimage" = "$REPO" ] || { echo "the manifest is for $vcomponent / $vimage, not $ROLE / $REPO; refusing to install" >&2; return 1; }
  case "$vdigest" in sha256:*) ;; *) echo "the manifest has no usable digest" >&2; return 1 ;; esac
  [ "${#vdigest}" -eq 71 ] && [ -z "$(printf '%s' "${vdigest#sha256:}" | tr -d '0-9a-f')" ] || { echo "the manifest digest is malformed" >&2; return 1; }
  IMAGE="$REPO@$vdigest"
  pull_quiet "$IMAGE" || { echo "cannot pull $IMAGE" >&2; return 1; }
  echo "verified release $vtag (CI-signed manifest, $ref)"
}
if [ -z "$IMAGE" ]; then
  if [ -z "$TAG" ]; then
    resolve_verified || exit 1
  else
    pull_quiet "$REPO:$TAG"
    if [ "$ENGINE" = podman ]; then DIGEST="$("$ENGINE" image inspect --format '{{.Digest}}' "$REPO:$TAG")"; IMAGE="$REPO@$DIGEST"
    else IMAGE="$("$ENGINE" image inspect --format '{{index .RepoDigests 0}}' "$REPO:$TAG")"; fi
    echo "warning: $REPO:$TAG was not signature-checked (--tag)" >&2
  fi
fi
echo "image: $IMAGE"

# 2. 卷。辅助容器以镜像自带的 app 用户（1654）读写，所以卷里的文件属主总是对的
"$ENGINE" volume inspect "$V_CFG" >/dev/null 2>&1 || "$ENGINE" volume create "$V_CFG" >/dev/null
"$ENGINE" volume inspect "$V_STATUS" >/dev/null 2>&1 || "$ENGINE" volume create "$V_STATUS" >/dev/null
in_volume() {
  "$ENGINE" run --rm -i --pull=never --network none --cap-drop ALL --security-opt no-new-privileges \
    -v "$V_CFG:/etc/moshiro" --entrypoint /bin/sh "$IMAGE" -c "$1"
}
[ "$ROLE" = backend ] || in_volume 'mkdir -p /etc/moshiro/update' </dev/null

# 3. 配置：身份由节点自己生成（Agent 和 Probe 一样），注册之后面板只认这把公钥
if [ "$ROLE" = backend ]; then
  # Backend 的环境文件放在宿主机上（单元用 --env-file 读取）；已有的不会被覆盖
  install -d -m 755 /etc/moshiro
  if [ ! -f /etc/moshiro/backend.env ]; then
    (
      umask 077
      {
        echo "MoShiro__DataDirectory=/data"
        echo "MoShiro__UpdateChannel=$CHANNEL"
        echo "MoShiro__ImageNamespace=$NAMESPACE"
        # 没有默认管理员：第一个管理员由部署的人在首次打开的页面上自己填，认领码在后端日志里（--admin-password 只给自动化用）
        if [ -n "$ADMIN_PASSWORD" ]; then
          echo "MoShiro__InitialAdminUser=${ADMIN_USER:-admin}"
          echo "MoShiro__InitialAdminPassword=$ADMIN_PASSWORD"
        fi
        [ -z "$PUBLIC_URL" ] || echo "MoShiro__PublicUrl=$PUBLIC_URL"
        # 其他设置原样写入，例如 --env MoShiro__CorsOrigins=http://localhost:5173 --env MoShiro__SecureCookies=false
        [ -z "$EXTRA_ENV" ] || printf '%s' "$EXTRA_ENV"
      } > /etc/moshiro/backend.env
    )
    echo "config: created /etc/moshiro/backend.env"
  fi
else
  ID_DIR="/etc/moshiro/$ROLE-id"
  if ! in_volume "test -f $ID_DIR/node.json" </dev/null; then
    [ -n "$SERVER" ] && [ -n "$JOIN" ] && [ -n "$NAME" ] || { echo "first deployment needs --server, --join and --name" >&2; exit 1; }
    "$ENGINE" run --rm --pull=never --network host -v "$V_CFG:/etc/moshiro" --entrypoint /usr/local/bin/fleet-enroll "$IMAGE" \
      --server "$SERVER" --join "$JOIN" --name "$NAME" --identity-dir "$ID_DIR"
  fi
fi
if [ "$ROLE" = agent ]; then
  if ! in_volume 'test -f /etc/moshiro/agent.json' </dev/null; then
    [ -n "$SERVER" ] || { echo "agent.json is missing; pass --server to create it" >&2; exit 1; }
    printf '{"fleet":{"server":"%s","identity_dir":"/etc/moshiro/agent-id"},"listen":{"bind":"::"}}\n' "$SERVER" \
      | in_volume 'cat > /etc/moshiro/agent.json'
    echo "config: created agent.json (listen.bind ::)"
  elif in_volume 'grep -q "\"bind\" *: *\"0\.0\.0\.0\"" /etc/moshiro/agent.json' </dev/null; then
    # 旧配置只监听 IPv4：改成 ::（双栈，IPv4 照常可用）
    in_volume 'sed -i "s/\"bind\" *: *\"0\.0\.0\.0\"/\"bind\": \"::\"/" /etc/moshiro/agent.json' </dev/null
    echo "config: listen.bind 0.0.0.0 -> ::"
  fi
elif [ "$ROLE" = probe ]; then
  # 旧格式（面板生成的 probe.json：没有 fleet 段）备份之后按新格式重写
  if in_volume 'test -f /etc/moshiro/probe.json' </dev/null && ! in_volume 'grep -q "\"fleet\"" /etc/moshiro/probe.json' </dev/null; then
    in_volume 'mv /etc/moshiro/probe.json /etc/moshiro/probe.json.old' </dev/null
    echo "config: old probe.json moved to probe.json.old"
  fi
  if ! in_volume 'test -f /etc/moshiro/probe.json' </dev/null; then
    [ -n "$SERVER" ] || { echo "probe.json is missing; pass --server to create it" >&2; exit 1; }
    printf '{"fleet":{"server":"%s","identity_dir":"/etc/moshiro/probe-id"}}\n' "$SERVER" \
      | in_volume 'umask 077; cat > /etc/moshiro/probe.json'
    echo "config: created probe.json"
  fi
fi

# 4. 单元和更新器（保持原来的更新策略）
install -d -m 755 "$CONF_DIR"
printf 'IMAGE=%s\n' "$IMAGE" > "$CONF_DIR/$ROLE.image"
install -m 644 "$PACKAGING/systemd/$UNIT_SRC" "/etc/systemd/system/$UNIT"
[ "$ROLE" != backend ] || sed -i "s/:29181/:$PORT/" "/etc/systemd/system/$UNIT"
set -- "$ROLE" --engine "$ENGINE" --volume "$V_CFG" --status-volume "$V_STATUS" --helper-image "$IMAGE" \
  --image-repo "$REPO" --unit "$UNIT" --policy "$POLICY" --no-enable
[ "$ROLE" != backend ] || set -- "$@" --channel "$CHANNEL" --health-url "http://127.0.0.1:$PORT/healthz"
[ -z "$AUTHFILE" ] || set -- "$@" --authfile "$AUTHFILE"
sh "$PACKAGING/update/install-updater.sh" "$@"

# 5. 启动
systemctl daemon-reload
systemctl enable "$UNIT" "moshiro-update-$ROLE.timer" >/dev/null 2>&1
systemctl restart "$UNIT"
systemctl start "moshiro-update-$ROLE.timer"
if [ "$ROLE" = backend ]; then
  n=0
  until curl -fsS -o /dev/null "http://127.0.0.1:$PORT/healthz" 2>/dev/null; do
    n=$((n + 1)); [ "$n" -lt 30 ] || { echo "backend did not become healthy" >&2; systemctl status "$UNIT" --no-pager >&2 || true; exit 1; }
    sleep 2
  done
else
  sleep 5
fi
systemctl is-active "$UNIT"